fix: reduce the impact of Redis cluster intermediate states

pkg/controllers/rediscluster/rediscluster_controller.go

@@ -73,37 +73,40 @@
 	}
 
 	// Check if the cluster is downscaled
-	if leaderCount := k8sutils.CheckRedisNodeCount(ctx, r.K8sClient, instance, "leader"); leaderReplicas < leaderCount {
+	if leaderCount := r.GetStatefulSetReplicas(ctx, instance.Namespace, instance.Name+"-leader"); leaderReplicas < leaderCount {
 		if !(r.IsStatefulSetReady(ctx, instance.Namespace, instance.Name+"-leader") && r.IsStatefulSetReady(ctx, instance.Namespace, instance.Name+"-follower")) {
 			return intctrlutil.Reconciled()
 		}
-
-		logger.Info("Redis cluster is downscaling...", "Current.LeaderReplicas", leaderCount, "Desired.LeaderReplicas", leaderReplicas)
-		for shardIdx := leaderCount - 1; shardIdx >= leaderReplicas; shardIdx-- {
-			logger.Info("Remove the shard", "Shard.Index", shardIdx)
-			//  Imp if the last index of leader sts is not leader make it then
-			// check whether the redis is leader or not ?
-			// if not true then make it leader pod
-			if !(k8sutils.VerifyLeaderPod(ctx, r.K8sClient, instance)) {
-				// lastLeaderPod is slaving right now Make it the master Pod
-				// We have to bring a manual failover here to make it a leaderPod
-				// clusterFailover should also include the clusterReplicate since we have to map the followers to new leader
-				logger.Info("Cluster Failover is initiated", "Shard.Index", shardIdx)
-				if err = k8sutils.ClusterFailover(ctx, r.K8sClient, instance); err != nil {
-					logger.Error(err, "Failed to initiate cluster failover")
-					return intctrlutil.RequeueWithError(ctx, err, "")
+		if masterCount := k8sutils.CheckRedisNodeCount(ctx, r.K8sClient, instance, "leader"); masterCount == leaderCount {
+			logger.Info("Redis cluster is downscaling...", "Current.LeaderReplicas", leaderCount, "Desired.LeaderReplicas", leaderReplicas)
+			for shardIdx := leaderCount - 1; shardIdx >= leaderReplicas; shardIdx-- {
+				logger.Info("Remove the shard", "Shard.Index", shardIdx)
+				//  Imp if the last index of leader sts is not leader make it then
+				// check whether the redis is leader or not ?
+				// if not true then make it leader pod
+				if !(k8sutils.VerifyLeaderPod(ctx, r.K8sClient, instance, shardIdx)) {
+					// lastLeaderPod is slaving right now Make it the master Pod
+					// We have to bring a manual failover here to make it a leaderPod
+					// clusterFailover should also include the clusterReplicate since we have to map the followers to new leader
+					logger.Info("Cluster Failover is initiated", "Shard.Index", shardIdx)
+					if err = k8sutils.ClusterFailover(ctx, r.K8sClient, instance, shardIdx); err != nil {
+						logger.Error(err, "Failed to initiate cluster failover")
+						return intctrlutil.RequeueWithError(ctx, err, "")
+					}
 				}
+				// Step 1 Remove the Follower Node
+				k8sutils.RemoveRedisFollowerNodesFromCluster(ctx, r.K8sClient, instance, shardIdx)
+				// Step 2 Reshard the Cluster
+				k8sutils.ReshardRedisCluster(ctx, r.K8sClient, instance, shardIdx, true)
 			}
-			// Step 1 Remove the Follower Node
-			k8sutils.RemoveRedisFollowerNodesFromCluster(ctx, r.K8sClient, instance)
-			// Step 2 Reshard the Cluster
-			k8sutils.ReshardRedisCluster(ctx, r.K8sClient, instance, true)
+			logger.Info("Redis cluster is downscaled... Rebalancing the cluster")
+			// Step 3 Rebalance the cluster
+			k8sutils.RebalanceRedisCluster(ctx, r.K8sClient, instance)
+			logger.Info("Redis cluster is downscaled... Rebalancing the cluster is done")
+			return intctrlutil.RequeueAfter(ctx, time.Second*10, "")
+		} else {
+			logger.Info("masterCount is not equal to leader statefulset replicas,skip downscale", "masterCount", masterCount, "leaderReplicas", leaderReplicas)
 		}
-		logger.Info("Redis cluster is downscaled... Rebalancing the cluster")
-		// Step 3 Rebalance the cluster
-		k8sutils.RebalanceRedisCluster(ctx, r.K8sClient, instance)
-		logger.Info("Redis cluster is downscaled... Rebalancing the cluster is done")
-		return intctrlutil.RequeueAfter(ctx, time.Second*10, "")
 	}
 
 	// Mark the cluster status as initializing if there are no leader or follower nodes

pkg/k8sutils/cluster-scaling.go

@@ -15,12 +15,11 @@
 // ReshardRedisCluster transfer the slots from the last node to the first node.
 //
 // NOTE: when all slot been transferred, the node become slave of the first master node.
-func ReshardRedisCluster(ctx context.Context, client kubernetes.Interface, cr *redisv1beta2.RedisCluster, remove bool) {
+func ReshardRedisCluster(ctx context.Context, client kubernetes.Interface, cr *redisv1beta2.RedisCluster, shardIdx int32, remove bool) {
 	redisClient := configureRedisClient(ctx, client, cr, cr.ObjectMeta.Name+"-leader-0")
 	defer redisClient.Close()
 
 	var cmd []string
-	currentRedisCount := CheckRedisNodeCount(ctx, client, cr, "leader")
 
 	// Transfer Pod details
 	transferPOD := RedisDetails{
@@ -29,7 +28,7 @@
 	}
 	// Remove POD details
 	removePOD := RedisDetails{
-		PodName:   cr.Name + "-leader-" + strconv.Itoa(int(currentRedisCount)-1),
+		PodName:   cr.Name + "-leader-" + strconv.Itoa(int(shardIdx)),
 		Namespace: cr.Namespace,
 	}
 	cmd = []string{"redis-cli", "--cluster", "reshard"}
@@ -274,18 +273,17 @@
 }
 
 // Remove redis follower node would remove all follower nodes of last leader node using redis-cli
-func RemoveRedisFollowerNodesFromCluster(ctx context.Context, client kubernetes.Interface, cr *redisv1beta2.RedisCluster) {
+func RemoveRedisFollowerNodesFromCluster(ctx context.Context, client kubernetes.Interface, cr *redisv1beta2.RedisCluster, shardIdx int32) {
 	var cmd []string
 	redisClient := configureRedisClient(ctx, client, cr, cr.ObjectMeta.Name+"-leader-0")
 	defer redisClient.Close()
-	currentRedisCount := CheckRedisNodeCount(ctx, client, cr, "leader")
 
 	existingPod := RedisDetails{
 		PodName:   cr.ObjectMeta.Name + "-leader-0",
 		Namespace: cr.Namespace,
 	}
 	lastLeaderPod := RedisDetails{
-		PodName:   cr.ObjectMeta.Name + "-leader-" + strconv.Itoa(int(currentRedisCount)-1),
+		PodName:   cr.ObjectMeta.Name + "-leader-" + strconv.Itoa(int(shardIdx)),
 		Namespace: cr.Namespace,
 	}
 
@@ -365,8 +363,8 @@
 }
 
 // verifyLeaderPod return true if the pod is leader/master
-func VerifyLeaderPod(ctx context.Context, client kubernetes.Interface, cr *redisv1beta2.RedisCluster) bool {
-	podName := cr.Name + "-leader-" + strconv.Itoa(int(CheckRedisNodeCount(ctx, client, cr, "leader"))-1)
+func VerifyLeaderPod(ctx context.Context, client kubernetes.Interface, cr *redisv1beta2.RedisCluster, leadIndex int32) bool {
+	podName := cr.Name + "-leader-" + strconv.Itoa(int(leadIndex))
 
 	redisClient := configureRedisClient(ctx, client, cr, podName)
 	defer redisClient.Close()
@@ -391,8 +389,8 @@
 	return false
 }
 
-func ClusterFailover(ctx context.Context, client kubernetes.Interface, cr *redisv1beta2.RedisCluster) error {
-	slavePodName := cr.Name + "-leader-" + strconv.Itoa(int(CheckRedisNodeCount(ctx, client, cr, "leader"))-1)
+func ClusterFailover(ctx context.Context, client kubernetes.Interface, cr *redisv1beta2.RedisCluster, shardIdx int32) error {
+	slavePodName := cr.Name + "-leader-" + strconv.Itoa(int(shardIdx))
 	// cmd = redis-cli cluster failover  -a <pass>
 	var cmd []string
 	pod := RedisDetails{

pkg/k8sutils/statefulset.go

@@ -26,6 +26,7 @@
 
 type StatefulSet interface {
 	IsStatefulSetReady(ctx context.Context, namespace, name string) bool
+	GetStatefulSetReplicas(ctx context.Context, namespace, name string) int32
 }
 
 type StatefulSetService struct {
@@ -76,6 +77,17 @@
 	return true
 }
 
+func (s *StatefulSetService) GetStatefulSetReplicas(ctx context.Context, namespace, name string) int32 {
+	sts, err := s.kubeClient.AppsV1().StatefulSets(namespace).Get(ctx, name, metav1.GetOptions{})
+	if err != nil {
+		return 0
+	}
+	if sts.Spec.Replicas == nil {
+		return 0
+	}
+	return *sts.Spec.Replicas
+}
+
 const (
 	redisExporterContainer = "redis-exporter"
 )